Message-ID: <4A14639D.8090701@wifo.ac.at>
Date: 2009-05-20T20:10:05Z
From: Serguei Kaniovski
Subject: Linear constraints for constrasts
Dear List!
How can I define contrasts (design matrix) that can all be included,
i.e. which do not require a control category be dropped. My application
(see below) does not suggest a sensible control category. I am thinking
of constraining the (treatment) contrasts to sum up to zero and dropping
the constant term in the regression. Is this a good idea? If yes, how to
achieve this in R?
I am estimating a GLM for bilateral country data. Each observation in on
a pair of countries, e.g. GER_USA, GER_JAP, USA_JAP. I constructed the
following contrasts: d_GER, d_USA, d_JAP, which take the value of 1 when
the country is in the pair and 0 otherwise, i.e.
?Bilat?, ?d_GER?, ?d_USA?, ?d_JAP?
GER_USA, 1, 1, 0
GER_JAP, 1, 0, 1
USA_JAP, 0, 1, 1
These contrasts highlight the effect of having a given country in the pair.
Thank you for your help!
Serguei Kaniovski